ALBATROSS: Automated Li-ion Battery Testing

• ALBATROSS is a fully integrated robotic platform for automated Li-ion battery testing, combining high-throughput screening, precise coin-cell assembly, and electrochemical evaluation.
• It utilizes a 6-axis robotic arm with advanced pick-and-place operations and custom fixtures to achieve a 97.7% assembly yield and reliable performance metrics.
• The system accelerates battery research by generating high-quality datasets for machine-learning integration and materials optimization in electrolyte development.

The Automated Li-ion Battery Testing Robot System (ALBATROSS) is a fully integrated robotic platform for high-throughput screening, assembly, and electrochemical evaluation of Li-ion coin cells, designed to address the increasing need for rapid and reproducible experimental workflows in electrolyte and battery materials research. The system operates entirely within an argon-filled glovebox, automating the entire workflow from electrolyte formulation to coin-cell cycling and impedance measurement, and is capable of unattended processing of up to 48 cells per run. ALBATROSS exemplifies an advanced convergence of robotics, liquid handling, precision assembly, and programmable orchestration for generating large, high-quality datasets essential for accelerating electrolyte and battery development (Lee et al., 15 Dec 2025).

1. System Architecture and Hardware Integration

The ALBATROSS platform incorporates a 6-axis robotic arm (xArm6, UFactory) with a 700 mm reach, installed inside a four-port glovebox to maintain an inert argon atmosphere, which is necessary for non-aqueous lithium battery chemistries. The robot accesses multiple stations: an electrolyte formulation unit, stacked assembly fixtures and posts, a crimper for hermetic sealing, and dual electrochemical testing gantries. The component-handling module uses a hybrid gripper, enabling the precise manipulation of diverse cell parts (thin foils, springs, spacers) and minimizing mechanical complexity. The bespoke fixtures, including 3D-printed assembly posts and rotary-indexing tables, are engineered for the sequential presentation of 48 complete coin-cell kits within the robot’s operating envelope (Figs. 2, S1–S4 in (Lee et al., 15 Dec 2025)). The electrolyte is formulated via an Opentrons OT-2 pipetting robot, equipped with a heated (60 °C) solvent reservoir to ensure solubility for high-EC content formulations and execute reproducible droplet-removal routines.

Coin-cell assembly is driven by a robot-executed pick-and-place sequence, encompassing the following ordered stack: 2032 can, lithium anode foil, separator, dosed electrolyte, NCM811 cathode, steel spring, spacer, and cap. The cell stack is then transferred to a motorized crimper (MSK-160E), which has an elevated Z-axis to permit unhindered robot placement. Subsequent to crimping, cells are shuttled to custom “cycling jigs” and “EIS jigs” for potentiostat connection (Neware CT-4008T, Biologics SP-150e), with two independent gantries each housing 24 parallel cycling channels and a dedicated EIS module (Lee et al., 15 Dec 2025).

2. Software Framework and Automated Control

ALBATROSS operation is managed by a centralized Omron NX102-9000 PLC, which sequences stepper motors, actuators, pneumatic controls, and all peripheral instrument signals. High-level logic and scheduling are implemented in Python as a state-machine workflow (Fig. 3 in (Lee et al., 15 Dec 2025)), encompassing the steps: “Formulate Electrolyte,” “Move Components,” “Assemble & Crimp,” “Transfer to Cycling/EIS,” “Conduct Cycling/EIS,” and “Data Acquisition & Save.” Instrument-level communications leverage custom drivers, enabling real-time data streaming from cycling hardware and impedance modules.

Crucially, coordinate calibration routines are streamlined: the system maps 336 potential pick-place points onto 8 master fixture references, which enables rapid recalibration—an essential feature for environments where regular maintenance (tip changes, fixture wear) can introduce spatial drift. Liquid-handling precision is ensured by automated gravimetric checks, with correction factors applied in software, and pipette droplet-removal (tip-touch) routines that reduce volumetric error.

3. End-to-End Workflow and Throughput

3.1 Electrolyte Formulation

Each coin cell receives precisely 70 µL of formulated electrolyte, with the pipetting module achieving ±0.5 µL accuracy. Typical compositions involve blending concentrated solutions and pure solvents, followed by 20 automated pipette mixing strokes over 3 minutes to ensure homogeneity. Heated reservoirs maintain EC above its crystallization point. Batch calibration via gravimetry calibrates each pipetting channel, maintaining consistency across runs.

3.2 Coin-Cell Assembly

The robot autonomously executes the full assembly stack in eight sequential steps per cell, achieving a total assembly yield of 97.7% (85 successes out of 87 attempts) with a mean cycle time of approximately 4 minutes per cell. Thus, 48 cells are completed in about 200 minutes per batch.

3.3 Electrochemical Evaluation

Formed cells undergo two initial galvanostatic charging cycles at 0.1 C, followed by 50 main cycles at 1 C (CC–CV charge, CC discharge, 3.0–4.2 V window). For impedance analysis, EIS spectra are measured after cycling at multiple C-rates (0.5–3 C) over 200 kHz to 0.1 Hz; each EIS module processes cells alternately for maximal parallelization. The overall system can assemble and initiate ~144 cells in an 8-hour shift, with cycling and EIS throughput bottlenecked by the ~6-day duration of full performance testing for each batch. In steady state, the platform yields ~240 cells/month with full cycling and impedance datasets (Lee et al., 15 Dec 2025).

4. Data Structure and Use Cases

Each cell yields high-fidelity records including full electrolyte formulation metadata (e.g., component concentrations and ratios), time-resolved cycling data (voltage, current, capacity, cycle count), and EIS spectra with fitted equivalent circuit parameters (R₁–R₄, CPEs). Traceability is maintained through comprehensive cell-ID labeling, environmental logs, and standardized reporting (Fig. 1, (Lee et al., 15 Dec 2025)). The platform uses a hierarchical file system or SQL-based schema, with electrolyte recipes, cell cycling data, and EIS results stored under unique identifiers. This infrastructure enables the deployment of machine-learning algorithms for predictive modeling of capacity retention, cycle life estimation, and active learning for materials discovery. ALBATROSS datasets are also compatible with interfacing to first-principles (DFT/MD) predictions, enabling closed-loop integration of computational and experimental workflows (Lee et al., 15 Dec 2025).

5. Performance Metrics and Reliability

Extensive validation demonstrates high reproducibility. Discharge capacities (N=40 per set) exhibit a relative standard deviation (RSD) of 1.040–1.210% at formation and 50th cycle, with corresponding values for manual assembly at 0.955–1.142%. EIS resistance fits across 45 cells show standard deviations of <3 Ω for charge-transfer (R₄) and SEI (R₃) components; bulk and contact resistances (R₁, R₂) have higher RSD, reflecting greater environmental and contact variability. The system is robust to pipetting and alignment errors via built-in tip-cleaning, regular gravimetric calibration, and precisely engineered 3D-printed posts. Fully inerted operation mitigates moisture/oxygen-induced error. The only significant downtime arises from routine replenishment of consumables (pipette tips, solvents, fresh components) (Lee et al., 15 Dec 2025).

Metric	Mean Value (μ)	Standard Deviation (σ)	RSD (%)
Formation discharge capacity	207.9 mAh/g	2.163 mAh/g	1.040
50th cycle discharge capacity	147.7 mAh/g	1.787 mAh/g	1.210
EIS charge-transfer R₄	39.39 Ω	1.462 Ω	3.71

Table: Representative performance metrics for ALBATROSS-assembled cells (Lee et al., 15 Dec 2025).

6. Limitations, System Variants, and Prospective Optimizations

Current constraints include restriction to 2032 coin-cell format and limited EIS parallelism (2/48 for full EIS spectra per batch). The long cycle time (≈6 days for cycling/EIS) is a major throughput bottleneck; implementing accelerated calendar or pulse protocols, or multiplexed EIS expansion, would increase productivity. The glovebox’s finite volume bounds the number and scale of liquid-handling and sensing modules; expansion would require substantial redesign.

Potential enhancements include modular gripper end-effectors for pouch or cylindrical cells, integration of in-line vision QC for electrode placement verification, expansion of EIS channels via multiplexers, and real-time AI-based adaptive scheduling to optimize test durations and cell assignments. In-line addition of viscosity, density, Raman or IR spectroscopy, or on-the-fly mass spectrometry, would facilitate richer data acquisition (Dave et al., 2019, Dave et al., 2021).

This suggests that the ALBATROSS principles extend directly from prior aqueous and non-aqueous electrolyte optimization robots such as Otto (Dave et al., 2019) and Clio (Dave et al., 2021), which pioneered automated formulation, measurement, and Bayesian-optimization-driven exploration, but generally stopped short of fully automated cell assembly or high-throughput cell testing. The modular, REST API-driven architecture and tight ML–robotics integration exemplified by those platforms are recommended blueprints for further ALBATROSS development.

7. Relationship to Automated Experimentation and Machine Learning-Driven Chemistry

ALBATROSS represents a direct application of the closed-loop, laboratory automation paradigm that couples robotic sample preparation, high-throughput measurement, and data-driven optimization, as established by systems such as Otto (Dave et al., 2019) and Clio (Dave et al., 2021). Otto’s hardware was optimized for aqueous precursor blending and electrochemical window discovery, using Dragonfly’s Gaussian process (GP)-based Bayesian optimization over multidimensional composition grids and an asynchronous API interface. Clio extended this paradigm to non-aqueous systems, incorporating high-precision dosing, in-line analytical modules, and active-learning-driven conductivity optimization. In both cases, black-box objectives (e.g., electrochemical window, conductivity) are iteratively maximized via acquisition functions such as Expected Improvement (EI) or multi-objective extensions (e.g., batch, qEI, multi-fidelity BO).

Plausible implications are that ALBATROSS, by incorporating complete cell-assembly and device-level testing, enables not just electrolyte discovery but direct device-performance-driven materials optimization. High-quality, annotated datasets from ALBATROSS support training of surrogate models (GPs, neural networks) for property prediction, can be leveraged in active learning schemes to accelerate discovery via experiment selection, and facilitate cross-validation with computational predictions, thus addressing key bottlenecks in battery innovation cycles (Lee et al., 15 Dec 2025).

---

References:

• "ALBATROSS: A robotised system for high-throughput electrolyte screening via automated electrolyte formulation, coin-cell fabrication, and electrochemical evaluation" (Lee et al., 15 Dec 2025)
• "Autonomous discovery of battery electrolytes with robotic experimentation and machine-learning" (Dave et al., 2019)
• "Autonomous optimization of nonaqueous battery electrolytes via robotic experimentation and machine learning" (Dave et al., 2021)